BRCA1-, BRCA2-, and PALB2-related Fanconi anemia: Scope to expand disease phenotypic features and predict breast cancer risk in heterozygotes

Summary

The recessive Fanconi anemia (FA) phenotype is used to classify BRCA1 (FANCS), BRCA2 (FANCD1), and PALB2 (FANCN) variants with respect to dominant hereditary breast-ovarian cancer syndrome. We assessed its utility by examining the phenotypic spectrum observed in individuals with bi-allelic BRCA1, BRCA2, or PALB2 pathogenic variants and exploring the relationship between cancer presentation and allele severity score based on variant molecular features. A data collection instrument comprising 158 Human Phenotype Ontology (HPO) terms was used to document clinical features for individuals with FA from published and/or prospectively collected sources (total n = 172, 43 previously unpublished). Distinct FA-related variants (15 BRCA1, 123 BRCA2, and 22 PALB2) were annotated for predicted molecular impact, location, observed splicing or functional impact, and potential in-frame splicing rescue and used to assign different permutations of allele severity scores, which were assessed for correlation with FA presentation features. The association of BRCA1 and BRCA2 allele severity score with the magnitude of breast cancer risk in heterozygotes was evaluated using case-control analysis. Clinical features extended beyond the HPO list, including 84 terms related by hierarchy and 94 additional terms. The BRCA2 genotype severity score was associated with age at cancer diagnosis in individuals with FA (p = 1.8 × 10⁻⁸). A similar permutation approach revealed significant differences in the magnitude of breast cancer risk according to the BRCA1 and BRCA2 allele severity score in heterozygotes. Our findings indicate the potential to redefine FA ORPHA:84 HPO terms and to use an allele severity scoring approach to predict cancer risk in individuals with bi-allelic or heterozygous BRCA1 or BRCA2 variants.

Graphical abstract

The recessive Fanconi anemia phenotype is used to classify BRCA1, BRCA2, and PALB2 variants with respect to dominant hereditary breast-ovarian cancer syndrome. Phenotype-genotype analysis of 178 individuals bi-allelic for BRCA1, BRCA2, or PALB2 pathogenic variants identified the potential to redefine ORPHA: 84 clinical features and predict the magnitude of cancer risk in heterozygotes.

Introduction

Fanconi anemia (FA) is a rare genetic disease characterized by chromosomal instability due to the defective function of proteins involved in DNA repair.¹ Of the 23 genes implicated in FA, one (FANCR [RAD51] [MIM: 617244]) causes autosomal-dominant FA, another (FANCB [MIM: 300514]) causes X-linked FA, and the remaining 21 have an autosomal-recessive pattern of inheritance where bi-allelic pathogenic variants cause disease.² In some cohorts, over 90% of individuals with FA are associated with recessive inheritance of pathogenic variants in FANCA (MIM: 227650), FANCC (MIM: 227645), FANCD2 (MIM: 227646), or FANCG (MIM: 614082). Rarer presentations of FA include bi-allelic inheritance of pathogenic variants in FA pathway genes known to increase risk of breast and/or ovarian cancer, including BRCA1 (FANCS) (MIM: 617883), BRCA2 (FANCD1) (MIM: 605724), PALB2 (FANCN) (MIM: 610832), RAD51C (FANCO) (MIM: 613390), and BRIP1 (FANCJ) (MIM: 609054).^2,3 The FA pathway is involved in the repair of DNA interstrand crosslink repair, with these breast-ovarian cancer susceptibility genes playing a key role in the homology-directed repair of DNA double-strand breaks that arise during interstrand crosslink removal.^2,3

The FA phenotype includes physical abnormalities that variably affect several parts of the body and an increased risk of developing progressive bone marrow failure from the first decade of life.^2,4 Individuals with FA also have an extraordinarily higher risk of early-onset solid and hematological cancers compared to the general population, as well as increased sensitivity to DNA crosslinking agents, such as platinum chemotherapeutics,^2,4 and ionizing radiation.⁵

The presence of a variant in trans with another known pathogenic variant in the context of a recessive phenotype is used as evidence toward variant pathogenicity, including the classification system of the American College of Medical Genetics and Genomics and Association of Molecular Pathology (ACMG/AMP).⁶ As such, it has been included as an evidence type in the ClinGen Variant Curation Expert Panel (VCEP) specifications for BRCA1, BRCA2, and PALB2 (https://cspec.genome.network/cspec/ui/svi/). Recommendations for defining the FA phenotype included phenotypic features for FA compiled for all FA genes combined, as drawn from GeneReviews,² and cancer diagnosis age cutoffs based on a literature review for individuals with FA bi-allelic for pathogenic variants in these genes specifically.⁷

There is evidence that the FA phenotype due to pathogenic variants in BRCA1, BRCA2, and PALB2 may be more severe (if not mostly embryonic lethal) compared to those that are due to pathogenic variants in most other FA genes. Individuals with FA due to bi-allelic BRCA2 or PALB2 pathogenic variants have a higher probability of developing cancer, and at much younger ages, compared to other, more commonly observed FA subtypes, requiring more intensive surveillance and screening.^8,9,10 The cancer spectrum is also unique, with embryonal tumors predominating, such as medulloblastoma.⁴ This is secondary to homology-directed repair being essential for DNA repair outside of the FA pathway. In FA mouse models, bi-allelic null variants in Brca1, Brca2, and Palb2 (and also Rad51C and Rad51) are embryonic lethal (reviewed in Guitton-Sert et al.¹¹). Bi-allelic BRCA1 pathogenic variant status was initially considered to be embryonic lethal, but there is emerging evidence that specific combinations of alleles support viability because they retain sufficient function and/or their deleterious impact is rescued through alternative splicing (reviewed in Hughes et al.¹²). Evidence for prenatal selection against specific bi-allelic combinations of alleles also exists for BRCA2. There are no reported individuals bi-allelic for Ashkenazi Jewish founder BRCA2 pathogenic variants; previous analysis showed a depletion of individuals with FA who harbor bi-allelic pathogenic variants within the largest BRCA2 exon 11, which harbors almost half of the reported pathogenic variants in this gene.⁴ There is also evidence to support that some alleles identified in individuals with FA due to bi-allelic BRCA2 variants may be hypomorphic in relation to experimentally measured variant impact on splicing or protein function or may be rescued through alternative splicing.^13,14,15,16

Here, we undertook a study to document the phenotypic and genotypic features of individuals with FA with bi-allelic pathogenic variants in BRCA1, BRCA2, or PALB2, to better inform the use of the FA phenotype to predict the pathogenicity of individual FA-detected alleles in the context of dominantly inherited cancer due to these three genes.

Subjects and methods

The collation and analysis of de-identified data for the overall study was approved by the QIMR Berghofer Human Research Ethics Committee (P1051). In relation to individuals with FA, data collection performed by external sites for prospective data analysis was approved under the following institutional review board (IRB) codes: Utrecht University METC 2022-3494 (E.G.-G.); NCI IBMFS study ClinicalTrials.gov: NCT00027274 (L.J.M. and N.G.); the Rockefeller University AAU-0112 (A.D.A., A.S., and T.B.); the Spanish Biobank of DNA Repair Syndromes collection no. C.0002193 (J.S. and R.P.); ICMR-National Institute of Immunohaematology NIIH/IEC/23-2019 (B.R.V. and M.G.); Instituto Nacional de Pediatría INP 053/2020, IRB: IRB00013674 (S.F., B.G.-d.-T., and M.Ó.F.-R.); Baylor College of Medicine H-41191 (S.L.); protocol #376800 (Penn) with a HIPAA waiver of consent (K.N.M., K.N., and S.D.); the Cincinnati Children’s Hospital Medical Center, IRB ClinicalTrials.gov: NCT02143830 (A.N.); the German Cancer Predisposition Syndrome Registry, DRKS00017382 (C.P.K.); and the Memorial Sloan Kettering Cancer Centre Institutional Review Board (J.A.K., M.I.C., and M.W.). All individuals included in prospective data collection provided written informed consent for de-identified data collection.

In relation to breast cancer case-control datasets, de-identified information was accessed as follows: BRIDGES data via the Breast Cancer Association Consortium (BCAC), concept ID 707; UK Biobank (UKB) data via the UKB Resource under application number 102655; and Cancer Risk Estimates Related to Susceptibility (CARRIERS) data via collaboration with the CARRIERS Coordinating Committee. Data and biospecimen collection for all participants contributing to the studies comprising these consortia were conducted with informed consent from participants.

Overview of study design

A schematic of the overall study design and workflow is shown in Figure 1.

Figure 1.

Schematic of overall study design and workflow

See subjects and methods and supplemental methods for further details on data collection, annotation, and analysis.

The FA dataset used for descriptive and other analyses comprised two partly overlapping groups: (1) a “retrospective dataset” curated from published reports of individuals with FA who harbor bi-allelic BRCA1, BRCA2, or PALB2 pathogenic variants and (2) a “prospective dataset” of individuals with FA who harbor bi-allelic BRCA1, BRCA2, or PALB2 variants obtained via collaboration that involved systematic clinical data collection for a subset of previously published individuals, as well as for previously unpublished individuals.

Retrospective data collection

Published reports of individuals diagnosed with FA who harbored BRCA1, BRCA2, or PALB2 variants were identified using the keywords “Fanconi anemia”/“Fanconi anaemia” and “FANCS,” “BRCA1,” “FANCD1,” “BRCA2,” “FANCN,” or “PALB2” in PubMed. Retrospective data collection was restricted to reports published from January 2000 to August 2024. References from within these publications were also reviewed for additional information relevant to individuals. We extracted, where available, genotype, sex, and age at FA diagnosis; results from chromosomal breakage studies; phenotypic features consistent with FA; cancer diagnoses and outcomes (specifically age of death); pathology findings, including toxicities from chemotherapy or radiation treatments and progressive bone marrow failure unrelated to cancer treatment; stem cell transplantation data; family cancer history; and reported sequence variants in genes of interest. We also contacted study authors to clarify variant details if required and invited them to participate in the prospective data collection arm of the study.

BRCA1, BRCA2, and PALB2 variant information was corrected to reflect Human Genome Variation Society (HGVS) nomenclature. Variants were first reviewed for evidence of pathogenicity using assertions (and related justifications) in ClinVar records as of January 2024. For variants considered to be of uncertain clinical significance (VUSs) after this initial review or variants absent from ClinVar, evidence for pathogenicity was assessed using ClinGen ENIGMA BRCA1 and BRCA2 VCEP specifications v.1.0 and ClinGen PALB2 VCEP specifications v.1.1. Following curation for variant pathogenicity and assessment for any reported FA phenotypic features, we excluded 29 individuals identified from publications (data_type code PE) and another five individuals from publications who remained ineligible after receiving updated information from collaborators (data_type code PUE); see Table S1 for a justification of the exclusions and further details. For another 30 individuals identified from the literature, the data from the publication were coded as superseded (data_type code PS) due to additional updated information provided as part of the detailed prospective collection (data_type code PU). Lastly, two individuals were identified through subsequent contact with the authors to be the same individual, with some resolution of additional characteristics (data_type annotation PE, overlap).

The retrospective dataset (data_type codes P and PS) included 129 individuals reported to have FA and variants in BRCA1 (n = 10), BRCA2 (n = 104), and PALB2 (n = 15) and (1) had sufficient details to assign HGVS nomenclature; (2) were bi-allelic for (likely) pathogenic variants in the same gene, with the exception of two individuals with one (likely) pathogenic variant and a second variant considered after review to be a VUS with high suspicion of pathogenicity (VUS suspicious; see Table S1 for further information); and (3) had a phenotype broadly consistent with FA diagnosis based on existing definitions that had been documented for use in variant interpretation by the ClinGen ENIGMA BRCA1 and BRCA2 VCEP, drawn from GeneReviews (see supplemental methods).

Development of the prospective data collection instrument

Review of the reported features in the retrospective dataset was used to inform development of a comprehensive data collection instrument that included individual diagnostic and demographic data, response to chemotherapy or radiation therapy, genotype, bone marrow transplantation, and clinical phenotype related to FA.

Human Phenotype Ontology (HPO) terms for phenotypes were obtained from the publicly available Orphanet rare disease database (https://www.orpha.net/en/disease), which is closely linked with the HPO project (https://hpo.jax.org/).¹⁷ HPO annotations and their expected frequencies within groups, defined as “very frequent” (80%–99%), “frequent” (30%–79%), or “occasional” (5%–29%) based on the published literature, were accessed on June 8, 2021, for FA (ORPHA:84) and for the differential diagnosis vertebral anomalies, anal atresia, congenital cardiac disease, tracheoesophageal fistula, renal anomalies, and limb defects with hydrocephalus (VACTERL-H; ORPHA:3412). A total of 106 HPO terms for FA (ORPHA:84; 13 very frequent, 10 frequent, and 83 occasional) and 33 HPO terms for VACTERL-H (ORPHA:3412; 12 very frequent, four frequent, and 17 occasional) were obtained from Orphanet. Eleven FA phenotypes overlapped with VACTERL-H; six of the 22 unique VACTERL-H terms were classified as very frequent, four were frequent, and 12 were occasional. Review of 42 potential additional FA-related terms identified through unpublished observations from collaborators or publications^18,19,20 revealed 30 more HPO terms not listed among the ORPHA:84 and ORPHA:3412 Orphanet terms. The resulting 158 HPO terms from all sources and their overlap are shown in Figure S1 and Table S2.

Five of the ORPHA:84 terms, HP:0003220 (abnormality of chromosome stability), HP:0002863 (myelodysplastic syndrome), HP:0002664 (neoplasm/cancer diagnosis), HP:0000478 (abnormality of the eye), and HP:0001871 (abnormality of blood and blood-forming tissues) are broad phenotypic abnormalities and were collected in more detail or as discrete phenotypic categories. We reviewed the remaining 153 specific terms for redundancy and mapped them according to the HPO hierarchy for parent-child-grandchild relationships based on HPO releases available as of April 5, 2023. A term was designated “parent” if at least one term on the list was its direct descendant, the latter designated as “child,” and designated “grandchild” if a term was a direct descendant of the child term (Table S2). We identified 22 parent terms, 41 child terms, and three grandchild terms within the list of 153 terms. The finalized list of HPO terms, including annotations for parent-child-grandchild relationships, was collated across 14 broad phenotypic abnormalities for prospective data collection, as shown in Table 1, together with the distribution of terms according to source. The data collection instrument, as provided to collaborators, is shown in Table S3. Each phenotypic category included an “other” free-text field, where submitters could describe additional phenotypes they considered as not listed among the existing terms.

Table 1.

Distribution of HPO terms in the prospective data collection instrument according to phenotypic category and source

Phenotypic abnormalities (categories)	aFA only	FA and VACTERL-H	FA and other sources	FA, VACTERL-H, and other	VACTERL-H only	VACTERL-H and other	Other sources only	Total
Growth	3	1	0	0	0	0	0	4
Limbs and musculoskeletal	15	0	1	1	5	0	6	28
Hair, skin, and nails	4	0	0	0	0	0	0	4
Blood and blood-forming tissues	4	0	0	0	0	0	1	5
Head or neck	10	0	3	1	0	0	5	19
Nervous system	6	3	0	0	2	0	1	12
Eye	8	1	2	0	3	0	1	15
Ear	3	0	0	0	3	0	1	7
Digestive	5	1	0	0	3	0	5	14
Genitourinary	11	1	3	1	2	0	9	27
Cardiovascular	9	0	0	0	1	0	0	10
Respiratory	0	1	0	0	1	0	0	2
Prenatal/birth	1	0	0	0	2	0	0	3
Endocrine	2	0	0	0	0	0	1	3
Total	81	8	9	3	22	0	30	153

Other sources are from collaborators and review of publications. HPO, Human Phenotype Ontology; FA, Fanconi anemia (Orphanet code ORPHA:84); VACTERL-H, vertebral anomalies, anal atresia, congenital cardiac disease, tracheoesophageal fistula, renal anomalies, and limb defects with hydrocephalus (Orphanet code ORPHA:3412).

^aFA-only terms listed exclude the following five ORPHA:84 terms: HP:0003220 (abnormality of chromosome stability), HP:0002863 (myelodysplastic syndrome), HP:0002664 (neoplasm/cancer diagnosis), HP:0000478 (abnormality of the eye), and HP:0001871 (abnormality of blood and blood-forming tissues). These are very broad phenotypic categories and were collected in more detail or as discrete phenotypic categories on the prospective data collection instrument.

Prospective data collection

Data obtained for 78 individuals with FA using the data collection instrument were reviewed, following the same variant classification processes as for the retrospective dataset. Five individuals were excluded from analysis due to a lack of convincing evidence that one or both variants were (likely) pathogenic or because no clinical phenotype information was provided. The final prospective dataset included information for 73 individuals with FA; 43 were unpublished (one individual with BRCA1 variants, 40 individuals with BRCA2 variants, and two individuals with PALB2 variants), and 30 individuals with BRCA2 variants were previously published with updates provided (information about overlap is provided in Table S1). There were 178 phenotypes provided under the other free-text field on our data collection instrument. These were reviewed against existing HPO terms in the collection instrument, assigned appropriate HPO terms, and mapped for their relationship to existing and other newly identified terms. The HPO identifier, its relationship to prospective data collection terms, the source, and the number of times reported are shown for these 178 additional phenotypes in Table S4.

Analysis of phenotypic features

The distribution of the 178 other terms identified in prospective data collection and their relationship to existing terms were presented graphically, using PhenoScore, an artificial-intelligence-based phenomics framework that allows for an efficient rendering of the extensive hierarchical network and can facilitate the identification of potential novel phenotypic connections.²¹ To produce this visualization, Python 3.10 was used to process and display the HPO structure, referencing HPO v.2022-06-11.

The frequency of individual HPO terms observed in our prospectively collected data for individuals with FA was estimated and plotted separately for (1) FA (ORPHA: 84) frequency bins designated for these disorders (occasional, 5%–29%; frequent, 30%–79%; very frequent, 80%–99%); (2) all VACTERL-H (ORPHA:3412) terms; and (3) terms derived from published sources not included in ORPHA:84 or ORPHA:3412. Information was tabulated and is shown graphically for descriptive purposes (Table S5).

Comparison of phenotypes reported in published vs. prospective datasets

For the 30 individuals with FA with both published data and prospectively collected data, information extracted from publications was compared to that collected by phenotyping using the collection instrument, to document the number of additional features captured by prospective data collection for descriptive purposes (Table S6).

We then compared the frequencies of FA phenotypes according to phenotypic category for data extracted from publication(s) to those collected prospectively for all individuals with FA (Table 2). For this analysis, the information updates provided for 30 previously published individuals with FA were included only in the frequency summaries for the prospective dataset. Information was stratified by age at FA diagnosis of ≤5 years (including individuals identified prenatally) compared to >5 years, to align with cancer diagnosis cutoffs set for use of the PM3 code in the ClinGen VCEP specifications for BRCA1 and BRCA2. For reporting and interpretation purposes, the Haldane-Anscombe correction^22,23 was applied for comparisons where at least one cell contained zero, and p ≤ 0.05 was considered statistically significant for this descriptive exploratory analysis.

Table 2.

Frequency of individuals with FA characteristics in published vs. prospectively collected data according to age at FA diagnosis

Phenotypic abnormalities (categories)	Age at FA diagnosis ≤5 yearsa					Age at FA diagnosis >5 yearsa
	Publishedb(N = 93)		Prospectivec(N = 64)		p valued	Publishedb(N = 35)		Prospectivec(N = 9)		p valued
	N	%	N	%		N	%	N	%
Growth	68	73.1	52	81.3	0.2	13	37.1	6	66.7	0.1
Limbs and musculoskeletal	53	57.0	37	57.8	0.9	9	25.7	4	44.4	0.3
Hair, skin, and nails	58	62.4	43	67.2	0.5	15	42.9	7	77.8	0.06
Blood and blood-forming tissues	2	2.2	23	35.9	1.6 × 10−8	1	2.9	5	55.6	5.0 × 10−5
Head or neck	36	38.7	30	46.9	0.3	8	22.9	1	11.1	0.4
Nervous system	50	53.8	46	71.9	0.02	17	48.6	4	44.4	0.8
Eye	18	19.4	28	43.8	0.001	4	11.4	4	44.4	0.02
Ear	14	15.1	26	40.6	0.0003	5	14.3	4	44.4	0.05
Digestive	24	25.8	24	37.5	0.1	4	11.4	2	22.2	0.4
Genitourinary	32	34.4	31	48.4	0.08	13	37.1	4	44.4	0.7
Cardiovascular	18	19.4	18	28.1	0.2	1	2.9	2	22.2	0.04
Respiratory	3	3.2	4	6.3	0.4	0	0.0	1	11.6	0.07
Prenatal/birth	1	1.1	5	7.8	0.03	0	0.0	0	0.0	1.0
Endocrine	4	4.3	10	15.6	0.01	3	8.6	2	22.2	0.3
Cancer diagnosis ≤5 years	83	89.2	35	54.7	9.6 × 10−7	3	8.6	e1	11.1	0.8

^aAge at FA diagnosis ≤5 years includes individuals with FA diagnosed prenatally; individuals with unknown age at FA diagnosis are excluded. Cases reviewed and coded PE or PUE in Table S1 have been excluded.

^bPublished individuals are coded P and PS (superseded by prospective data collection) in Table S1, and are based on phenotype data from the original published reports. One individual with age at FA diagnosis unknown was excluded from published analyses.

^cProspective individuals are coded U (unpublished) and PU (published individuals updated by prospective data collection) in Table S1, and are based on data collected using the prospective data collection instrument.

^dp values where contingency tables included N = 0 are based on a Haldane correction of 0.5 applied to all cells.

^eThis individual was diagnosed with Wilms tumor at age 0.9 years, but the FA diagnosis was not made until age 6.5 years.

Assignment of ACMG/AMP PM3 code for individual variants

For each individual with FA considered eligible for phenotypic and genotype analyses (retrospective and prospective individuals combined), we reviewed all data collated against the characteristics specified to assign the PM3 code for recessive presentation following ClinGen VCEP specifications for BRCA1, BRCA2, and PALB2, i.e., chromosome test findings, physical features, pathology findings, and cancer diagnosis at age ≤ 5 years (supplemental methods). The rationale for this review was to (1) determine for how many individuals there was sufficient information to assign PM3 evidence toward pathogenicity for the alleles detected in these individuals with FA; (2) assess if there might be other features to be considered as important in assigning the PM3 code for FA due to bi-allelic pathogenic variants in these three genes; and (3) provide a resource for future classification in the context of VCEP activities, including designating which of multiple relatives should be selected as having the most severe presentation for PM3 code application.

Annotation of predicted and observed variant type and molecular impact for FA-related alleles

For the combined retrospective and prospective datasets (172 individuals), each distinct FA-related allele (n = 159) was annotated for likely mechanism of impact. As detailed in the supplemental methods, annotations included variant location (exon), variant type as a surrogate for primary molecular consequence (synonymous, intronic, missense, deletion, and protein termination codon [PTC]), predicted splicing impact, and experimentally determined variant impact as measured using protein functional and/or splicing assays. Exon numbering matched the exon descriptions of the MANE transcript for BRCA1 (GenBank: NM_007294.4) or BRCA2 (GenBank: NM_000059.4), i.e., sequential (not legacy) exon numbering was used for BRCA1. This included more detailed curation of observed/predicted spliceogenic variants located at the BRCA2 exon 1 donor motif (Figure S2) and the BRCA2 exon 2 acceptor and donor motifs (Figure S3). Aligning with VCEP specifications, variants encoding a PTC were also assessed for susceptibility to nonsense-mediated decay (NMD), and missense substitutions were annotated for location inside or outside a known functional domain. Annotations for distinct alleles are shown in Table S7. Each variant was then labeled with an annotation acronym capturing combined variant type, splicing prediction, impact based on experimental data (and level of impact where available), and NMD prediction for PTC variants (Table S8). Exon-level annotation was performed to assess the possibility of rescue of PTC alleles due to single-exon or multi-exon framing (Table S9).

In addition, the level and consequence of impact on splicing and/or the level of impact on function were tabulated for BRCA2 FA alleles, specifically to investigate the hypothesis that FA-related alleles may be enriched for hypomorphic variants (Table S10). For this analysis, we defined a hypomorphic allele as a variant with partial impact on mRNA level or protein function, altered mRNA transcript expression that is expected/shown to result in an incomplete impact on function at the protein level, or conflicting results across multiple assays capturing the impact on protein function. The initial descriptive analysis was performed using existing functional assay data collated for use as per BRCA2 VCEP specifications v.1.1, with secondary analysis considering results from another two recently published multiplexed assays of variant effect (MAVE) assessing function for variants in the BRCA2 DNA-binding domain.^24,25

In addition to the experimental impact annotation, a graphical representation of allele pairs was undertaken to inform allele severity annotation based on both allele effect and exonic location, separately for each gene. Bi-variant positional schematic plots were generated using R v.4.3.1 with the tidyverse (v.2.0.0), ggrepel (v.0.9.6), and RColorBrewer (v.1.1-3) packages. Final formatting was performed using Inkscape (v.0.92.3). Coordinates were plotted for a single individual with FA per family.

Annotation of allele and genotype severity score for assessing association with clinical presentation

Previous findings from Radulovic et al. showed that cancer age at diagnosis in individuals with FA due to bi-allelic pathogenic variants in BRCA2 was associated with allele severity impact, defined based on variant type, a location ≤ or > exon 11, or splicing prediction score changes for splice-site dinucleotide variants.²⁶ Building on this approach, we re-created an allele scoring system similar to that of Radulovic et al.²⁶ and also created our own BRCA2 allele severity scoring system, which included several permutations of severity score based on variant type, predicted/observed molecular impact, location, protein stability, and possible rescue due to in-frame exon skipping or possible upregulation of known alternative in-frame isoforms (Tables S7 and S9; see the supplemental methods for details). Using labels as defined by Radulovic et al., scores for individual alleles ranged from 0 (most severe impact) to 2 (least severe impact).

We then conducted Kaplan-Meier survival analysis and Cox regression modeling to investigate the impact of genotype severity score (i.e., allele 1 severity score + allele 2 severity score) on age at cancer diagnosis in individuals with FA due to BRCA2 variants. For this analysis, prenatal individuals with FA were excluded, as was a single cancer-affected individual with unknown age at diagnosis. Individuals affected with cancer were censored as affected at the age of diagnosis, and individuals with benign tumors or who were cancer unaffected were censored at the age of death or last known review. Survival analysis and Cox regression modeling were performed using STATA SE v.18 (Stata, USA). Table S1 provides more details on assumptions made regarding inferred age at presentation/review for a subset of individuals.

The best-fitting Cox regression model for the association of the BRCA2 genotype score with age at cancer diagnosis was selected for additional analysis that separated genotype score 2 according to the component allele severity scores (0+2 vs. 1+1). The allele severity scoring system that resulted in the best-fitting BRCA2 allele severity model was then applied to derive allele and genotype severity scores for individuals with FA due to variants in BRCA1 or PALB2, for descriptive analysis.

Case-control analysis estimating risk of breast cancer by allele severity score

Building on findings showing the association of combined allele severity score with age at cancer onset, we assessed the risk of breast cancer associated with heterozygous BRCA1 and BRCA2 pathogenic variant status, stratified by allele severity score assigned using the same approach as for FA-related alleles (see the supplemental methods for more details). Following the approach applied for Cox regression analysis of individuals with FA due to BRCA2 variants, analysis was performed considering different permutations of the allele severity score. Burden analyses were performed, in which odds ratios (ORs) and 95% confidence intervals (CIs) for breast cancer associated with the presence of any variant in a given allele severity score category were estimated by means of logistic regression in a combined dataset of 96,991 female breast cancer-affected individuals and 302,116 unaffected control subjects aggregated from the BRIDGES study of the BCAC,²⁷ the CARRIERS consortium,²⁸ and the UKB.^29,30 Dataset quality control and filtering criteria have been described elsewhere.³¹ Associations were adjusted for age and study country for the BCAC dataset, age and ethnic group for the CARRIERS dataset, and age and genetic ancestry for the UKB dataset. ORs and standard errors estimated from each dataset were combined in a fixed-effects, inverse-variance meta-analysis using the “metafor” R package to derive an overall test of association.

Results and discussion

Phenotypes observed in prospective data collection

Overall, 296 features were observed in individuals with FA included in the prospective cohort. The number of observations of each HPO term is shown in Table S4. The features observed included 82 of 104 ORPHA:84 terms, 14 of 22 terms originating only from the ORPHA:3412 VACTERL list, and 20 of 30 terms sourced from publications or collaborators. Of the 178 phenotypes provided as free text, simplistic relationship mapping established that 84 terms could be related by hierarchy to other features listed in the data collection instrument: 68 terms were descendants of existing terms (24 child, 29 grandchild, 11 great grandchild, and 4 great-great grandchild), two terms were grandparents of existing terms, and 15 terms were in the same HPO hierarchy as existing terms (designated “sibling” terms). The remaining 94 terms were considered to be newly identified terms not directly linked to HPO terms listed in the data collection instrument (called “new” terms for display purposes).

The distribution of other terms across phenotypic categories, and their relationship mapped to existing terms in the prospective data collection sheet, is shown in Figure 2. The most frequently reported other phenotypes were short thumb (also reported as thumb hypoplasia/brachydactyly, n = 10), failure to thrive/small for gestational age (n = 6), pelvic kidney (n = 6), posteriorly rotated ears (n = 4), stenosis of the external auditory canal (n = 4), and vesicoureteric reflux (n = 4). Ten terms (sacral dimple, preaxial polydactyly, bone marrow hypocellularity, convex nasal ridge, holoprosencephaly, bilateral sensorineural hearing impairment, fused or horseshoe kidney, ventricular septal defect, growth hormone deficiency, and small/ectopic pituitary) were each reported three times, and another 26 terms were each reported twice.

Figure 2.

Distribution of other Human Phenotype Ontology terms obtained in prospective data collection

The 178 other terms were distributed across phenotypic categories as follows: neoplasms (n = 2, specified by histology); growth (n = 3); limbs and musculoskeletal (n = 36); hair, skin, and nails (n = 14); blood and blood-forming tissues (n = 6); head or neck (n = 19); nervous system (n = 27); eye (n = 15); ear (n = 10); digestive (n = 11); genitourinary (n = 17); cardiovascular (n = 10); respiratory (n = 1); prenatal/birth (n = 3); and endocrine (n = 5).

These findings are apparent from the network visualization (Figures 3 and S4), showing that the newly identified terms can be ontologically connected to existing FA HPO terms, albeit with many newly identified terms presenting as outliers in the hierarchical connection (Figure S4). These findings indicate the need to redefine the existing list of FA-related HPO terms using an ontogenic phenomics approach and possibility to extend this list to include several newly identified features in the future.

Figure 3.

Overview of phenotypic data in Human Phenotype Ontology terms

This network visualization illustrates the distribution and relationships of phenotypic terms as defined by the Human Phenotype Ontology (HPO). Each circle (node) represents an HPO term, and edges (gray lines) indicate hierarchical or ontological connections among those terms, following the hierarchy of the HPO. Two key categories of terms are displayed. (1) Newly identified terms (orange nodes): these are phenotypic features previously not reported as related to Fanconi anemia and not directly linked to HPO terms in the data collection instrument. (2) Existing terms (blue nodes): well-established HPO terms and those related by hierarchy. Node size reflects the relative frequency of a particular phenotype in the dataset—larger circles denote phenotypes observed more frequently among the subjects studied. The legend lists the specific phenotypic features representing >20% of observations, with numbers within the node representing the most frequent (abnormality of chromosome stability) to the least frequent (aplasia/hypoplasia of fingers).

Frequency of ORPHA: 84 and ORPHA: 3412 HPO terms observed in the prospective dataset of individuals with FA due to variants in BRCA1, BRCA2, or PALB2

The HPO terms classified as very frequent (80%–99%) for FA (ORPHA:84) ranged in frequency in our prospective data from 0% to 81.2%, frequent (30%–79%) ranged from 1.4% to 68.5%, and occasional (5%–29%) ranged from 0% to 56.2% after correction for sex where relevant (Tables S5A–S5C). HPO terms unique to VACTERL-H (ORPHA:3412) that were classified as very frequent ranged in our dataset from 0% to 5.3%, whereas frequent ranged from 0% to 10.7%, and occasional ranged from 0% to 12% (Table S5D). HPO terms added from FA-related publications^18,19,20 and personal communications with collaborators ranged in frequency from 0% to 23.5% (Table S5E).

Comparison of phenotypic data reported in publications vs. prospective collection

For 27 of the 30 individuals with FA and for whom data were obtained from existing sources and via prospective collection, more features were provided through the standardized prospective collection compared to extraction from publications (Table S6). For several individuals, the additional phenotype data led to altered assignment of the PM3 code, for use in variant curation.

Compared to data extracted from published studies (Table 2; Figure S5), prospective reports for individuals with FA diagnosed at ≤5 years of age had significantly higher frequencies of abnormalities in Blood and blood-forming tissues (36% vs. 2%; p = 1.6 × 10⁻⁸), Nervous system (72% vs. 54%; p = 0.02), Eye (44% vs. 19%, p = 0.001), Ear (41% vs. 15%; p = 0.0003), Prenatal/Birth (8% vs. 1%; p = 0.03), and Endocrine (16% vs. 4%; p = 0.01). Conversely, published studies were significantly more likely than prospective studies to report individuals diagnosed with FA at ≤5 years of age who also had at least one cancer diagnosis (89% vs. 55%; p = 9.6 × 10⁻⁷). In the subset of individuals diagnosed with FA at >5 years of age, prospective reports had significantly higher frequencies of abnormalities of Blood and blood-forming tissues than published reports (56% vs. 3%; p = 5.0 × 10⁻⁵) and Eye (44% vs. 11%; p = 0.02), Ear (44% vs. 14%; p = 0.05), and Cardiovascular (22% vs. 3%; p = 0.04) abnormalities. These differences likely reflect biases in case studies identified as well as biases in the associated information selected for presentation in published reports of individuals with FA, e.g., there was no or poor annotation of non-cancer hematological features in most publications.

Alignment of recessive presentation features with VCEP specifications

Review of each individual with FA for the VCEP-specified characteristics designated to assign the PM3 recessive code is captured in Table S1. The 172 individuals with FA represented 140 different families. Of the nine individuals with prenatal presentation, three met the PM3 criterion, while characteristics reported for the remainder were insufficient to assign PM3—all presented with physical features of FA but were (apparently) untested for chromosomal breakage and had no information reported relevant to pathology findings.

For the remaining 163 individuals, PM3 was met by 146 individuals representing 121 families; considering only code assignment per family, the code was met by 107 individuals at moderate weight and 14 at supporting weight. For the remaining individuals, the PM3 code was not met largely because of insufficient information. After considering censoring age due to death, only three individuals had a presentation considered “not characteristic” as defined by the VCEP specifications: one individual with positive chromosome test results, a breast cancer diagnosis at age 22 years, and no reported physical or pathology features³²; one individual with negative chromosome test results and a colorectal cancer diagnosis at age 37 years³³; and one individual with negative chromosome test results, mild physical features, and toxicity associated with treatment for a breast cancer diagnosis at age 30.³⁴ Notably, the latter individual was bi-allelic for the established high-risk pathogenic allele BRCA1 c.181T>G (p.Cys61Gly) and the known pathogenic allele BRCA1 c.5096G>A (p.Arg1699Gln)³⁴ confirmed to be associated with reduced penetrance.^35,36

Overview of predicted and observed variant type and molecular impact for FA-related alleles

The combined dataset of non-overlapping individuals with FA identified from published or unpublished data, after exclusions, included 11 individuals with FA due to BRCA1 variants (from 9 distinct families), 144 individuals with FA due to BRCA2 variants (from 115 families), and 17 individuals with FA due to PALB2 variants (from 15 families) (Table 3). Many of the FA-related alleles were observed multiple times across different families. Specifically, 15 distinct BRCA1 alleles were observed in 9 families, 122 distinct BRCA2 alleles in 115 families, and 22 distinct PALB2 alleles in 15 families.

Table 3.

Distribution of FA-related alleles according to predicted and observed variant impact on function

	BRCA1	BRCA2	PALB2
Total number of individuals with FAa	11	144	17
Individuals with FA (published only)	10	74	15
Individuals with FA (published updated)	0	30	0
Individuals with FA (previously unpublished)	1	40	2
Total number of families	9	115	15

Variant annotation descriptionb	Distinct BRCA1 alleles	Allele observationsc	Distinct BRCA2 alleles	Allele observationsc	Distinct PALB2 alleles	Allele observationsc
Total n	15	18	122	230	22	30
Synonymous/intronic splicing predicted/proven	0	0	6	13	3	4
Missense/in-frame deletion within functional domain	6	6	19	29	0	0
Missense in domain splicing predicted and proven	0	0	2	3	0	0
Missense out domain splicing predicted and proven	0	0	4	15	0	0
Large deletiond	0	0	1	1	1	1
PTCd	8	11	69	109	10	13
PTC no NMDd	0	0	3	14	4	8
PTC splicing predicted/provend	1	1	3	23	3	3
Intronic splice-site dinucleotide predicted loss of function as per PVS1 decision treed	0	0	11	19	1	1
Intronic splice-site dinucleotide predicted/proven indel or regulatory	0	0	4	4	0	0
Proportion of variants with (predicted) impact considered loss of function as per ClinGen PVS1 decision treed	0.60	0.67	0.71	0.72	0.86	0.87

^aIncludes only individuals with minimum data to be included in the phenotype-genotype component of the study (P, PU, and U codes, Table S1).

^bSee the supplemental information and Table S7 for more details on variant annotation and categorization. A BRCA1 small in-frame deletion was coded as missense for presentation.

^cTotal observations across unrelated families (i.e., excluding relatives).

^dVariant types considered assessable via a PVS1 decision tree, as detailed in Abou Tayoun et al.³⁷

Alleles generally considered to be loss of function based on variant type (PTC, splice-site dinucleotide, and deletion, as assessed using a PVS1 decision tree³⁷) comprised the majority of distinct alleles across the three genes (BRCA1, 60%; BRCA2, 71%; and PALB2, 86%) and also the majority of all alleles observed per gene in the unrelated families (≥67%). This included three distinct BRCA2 and four distinct PALB2 PTC variants predicted to escape NMD but which would still result in loss of critical functional residues. The five BRCA1 distinct variants encoding a missense substitution, plus one in-frame deletion variant, all fell within a known clinically important functional domain (two in the RING domain and the remainder in the BRCA1 C-terminal domain). For BRCA2, 21 predicted missense substitution variants fell within the DNA-binding domain, two of which had a mechanism of impact through splicing; the remaining four missense variants located outside of a functional domain all impacted splicing.

A summary of experimental data indicating the level of impact on splicing and/or (protein) function for BRCA2 FA alleles is shown in Table S10, with a more detailed explanation of the findings reported in Note S1. In brief, experimental data were available for all BRCA2 missense and synonymous variants, all intronic variants located outside of the splice-site dinucleotide positions, and half of the intronic variants located at the dinucleotide positions. Excluding variants encoding a PTC, 16% (6/38) of those tested experimentally to capture the allele molecular effect (protein and/or splicing) showed a partial or conflicting functional or splicing impact based on one or more assays. In comparison, a review of calibrated functional assay results compiled for use in VCEP curation under the updated specifications v.1.2 (https://cspec.genome.network/cspec/ui/svi/doc/GN097) showed that impact was partial or conflicting for 29% (80/270) of the BRCA2 missense variants recorded as having any impact (partial or complete) from at least one study. This simplistic comparison suggested that the proportion of variants with a partial or conflicting impact on function vs. a complete impact is not obviously greater for BRCA2 alleles observed in individuals with FA compared to the pool of variants with assay results. Additionally, when considering findings from two recently published MAVEs capturing the functional impact at the mRNA and protein levels,^24,25 yet to be formally included in VCEP specifications, further inconsistencies were introduced. These discrepancies were due to a difference in functional impact for one or both of the MAVE studies compared to those drawn from existing functional data (24 variants) or differences in results between the two MAVE studies. Overall, this observation highlights a limitation in using “lack of agreement between functional assays” to infer whether a variant may potentially have partial impact on function, as different functional assays can measure different aspects of function and also inconsistency in functional impact results for individual variants may reasonably be expected to increase as more assay findings (and thus more experimental errors) are included in the annotations.

Within-gene distribution of bi-allelic pairs observed in individuals with FA

Graphical representation of allele pairs is shown in Figures S6 (BRCA1), S7 (BRCA2), and S8 (PALB2). The majority of BRCA1 pairs had one or both alleles located in exon 10, consistent with the hypothesis that upregulation of naturally occurring in-frame splicing may be a key factor acting to rescue lethality for BRCA1 bi-allelic pathogenic variant status.^12,38 The remaining three pairs included at least one missense variant, including the BRCA1 c.5096G>A (p.Arg1699Gln) variant previously reported to demonstrate ambiguous findings across different functional assays and confirmed to be associated with reduced breast and ovarian cancer risk compared to the average BRCA1 PTC pathogenic variant.^35,36 For BRCA2, the most notable observation was the paucity of bi-allelic variants in the largest exon (exon 11). Only one individual was bi-allelic for pathogenic variants in this exon, the homozygote for BRCA2 c.3751dup (p.Thr1251Asnfs^∗14), diagnosed after pregnancy termination at 13 weeks. For PALB2, it was notable that six of the 11 individuals with FA had at least one PTC allele predicted to escape NMD.

Association of BRCA2 genotype severity score with age at cancer diagnosis

Cox regression analysis was undertaken to assess the association between age at cancer diagnosis and genotype severity score (based on the summed allele severity score), expanding previously reported findings by Radulovic et al.²⁶ Using an allele scoring system aligned with that of Radulovic et al.,²⁶ there was a significant association with age at cancer onset (Figure S9, chi squared (4df) = 18.04, p = 0.001). The result was slightly more significant using the baseline severity score system selected for this study (Figure S10, chi squared (4df) = 19.15, p = 0.0007). Association results from baseline and several permutations of this severity scoring system are shown in Table S11. The best-fitting model (Figure 4A) was as follows: the severity score of missense variants was based on predicted protein instability rather than on functional assay data, a less severe score was applied to splice variants with a partial impact on splicing or resulting in large in-frame transcripts that do not delete a known functional domain, and a less severe score was applied to PTC variants located in exons with the potential for rescue due to single or multi-exon in-frame skipping. When the genotype severity score was categorized into five groups from most to least severe (0, 1, 2, 3, and 4), age at cancer onset was significantly later for individuals with a genotype score of 1 (hazard ratio [HR]: 0.46), 2 (HR: 0.23), 3 (HR: 0.13), and 4 (HR: 0.11) compared to those with a genotype severity score of 0 (chi squared (4df) = 40.94, log-rank test p = 2.8 × 10⁻⁸). The strength of association was similar (chi squared (3df) = 41.81, log-rank test p = 1.8 × 10⁻⁸) when considering the component allele severity scores that contributed to the genotype scores (Figure 4B). The HR was unchanged for allele score 0+1 (i.e., genotype severity score 1) and was marginally different (0.25) for the subset of individuals with genotype severity score 2 due to component allele scores 1+1. The HR was 0.13 for the pool of remaining genotype combinations, each comprising at least one allele score 2 (allele scores 2+0, 2+1, and 2+2).

Figure 4.

Kaplan-Meier probabilities and Cox regression analysis assessing the association of genotype severity score with age at diagnosis of cancer for BRCA2-related individuals with FA

(A) For best-fitting genotype severity score model, the allele severity scores were based on PTC (noNMD) score 2; PTC exon 10, 11, and 12 score 1 (exon rescue); PTC exon 4–7 score 1 (potential multi-exon rescue); PTC exon 14 score 1 (potential escape from NMD due to large exon size); other PTC score 0; splice-site dinucleotide with regulatory impact score 1; splicing complete impact and observed transcript expected to under NMD score 0; splicing complete impact and transcript encodes larger in-frame deletion expected to escape NMD score 1; splicing partial impact score 1; missense outside domain (all with splice impact) score 0; missense in domain predicted unstable score 1; and missense in domain predicted stable score 2. See Table S7 for further description of the annotations informing allele severity scores. Survival plots, Cox proportional hazard ratios, and their 95% confidence intervals (95% CI) are estimated using a genotype severity score of 0 as the reference for severity scores 1, 2, 3, and 4. p values are for the tests of the null hypothesis that each hazard ratio is 1.

(B) Considering allele score components of genotype severity, the allele severity scores were applied as for best-fitting genotype severity model shown in (A), but genotype severity was grouped to consider the component allele severity score as follows: genotype score 0, all based on allele severity 0+0; genotype score 1, all based on allele severity 0+1; genotype score 2, based on allele severity 1+1; and genotype score 2 or greater (score 2, 3, or 4), where at least one allele was coded with severity score 2 and the remaining allele was score 0 or 1 or 2. Survival plots, Cox proportional hazard ratios, and their 95% confidence intervals (95% CI) are estimated using an allele severity score combination 0 + 0 as reference. p values are for the tests of the null hypothesis that each hazard ratio is 1.

Association of BRCA1 and PALB2 genotype severity score with FA phenotype

The allele severity scoring system that yielded the best-fitting BRCA2 allele severity model was then applied to derive allele and genotype severity scores for individuals with FA due to BRCA1 or PALB2 pathogenic variants. Comparisons of the relative proportions of alleles by severity score for each gene are shown in Table S12. Although these comparisons were based on small numbers, it was notable that no BRCA1 alleles were assigned a severity score of 0 (i.e., PTC variants or splicing alleles with no rescue expected) and that most (72%) alleles were assigned score of 2, with most of these being PTC variants with potential rescue by in-frame isoform expression and potential escape from NMD due to large exon size (Table S7). This observation is consistent with the hypothesis that individuals bi-allelic for BRCA1 pathogenic variants can only survive if both variants retain some function. In contrast, PALB2 FA-related alleles were distributed across all three severity score categories, as already observed for BRCA2 FA-related alleles.

Clinical presentation age and features, grouped into broad categories for simplicity, showed no obvious pattern in relation to BRCA1 or PALB2 genotype severity scores or the component allele scores (Table S13). Genotype score distribution was consistent with the distribution of individual allele severity scores for the different genes. While most individuals with BRCA1-related FA (nine out of 11) had a genotype severity score of 3 or more, a genotype severity score of 2 was assigned for the single prenatal individual and for two individuals with relatively late age at cancer diagnosis (aged 23 and 30 years). Almost half (47%) of individuals with PALB2-related FA had a genotype severity score of 1, including the single prenatal individual. A genotype severity score of 2 was assigned to 42% of individuals with BRCA2-related FA, including four of seven individuals with prenatal presentation. Additionally, comparison of the phenotype for pairs or trios of siblings across 30 families (Table S14) revealed differences in age at cancer diagnosis (categorized as ≤ vs. > 5 years) for both of two families with BRCA1-related FA (with a genotype severity score of 4), one of two families with PALB2-related FA (with a genotype severity score of 1), and for three of 28 families with BRCA2-related FA (with genotype [allele] severity scores of 1 [0+1], 2 [0+2], and 3 [1+2]). There was some evidence to suggest that within-family variability in cancer phenotype severity (as categorized above) increased as the genotype severity score increased (p trend = 0.04).

Association of BRCA1 and BRCA2 allele severity score with risk of breast cancer

The results from a meta-analysis of three breast cancer case-control datasets are shown in Table 4, for different allele severity score options, with the results for each dataset detailed in Tables S15 (BRCA1) and S16 (BRCA2). For BRCA1, the baseline severity scoring option showed the best separation of effect (p_continuous = 7.11 × 10⁻¹⁵³, Table S15), with alleles annotated as having a severity score of 2 demonstrating a reduced breast cancer risk (OR = 4.88, 95% CI: 2.89–8.26; p = 3.34 × 10⁻⁹) compared to those with a severity score of 0 (OR = 11.36, 95% CI: 8.51–15.17) and a severity score of 1 (OR = 11.06, 95% CI: 8.86–13.80). Applying allele scoring options A, B, and C led to a few changes in allele severity coding for distinct alleles. In contrast, further upweighting PTC variants within in-frame exon 10 (option D) due to evidence for possible rescue by naturally occurring splicing changed the allele score annotation for a large proportion of variants/individuals, and there was a large overlap in CIs around the risk estimates for the three allele severity score categories. That is, of the allele scoring options assessed in this study, the baseline scoring approach provided the best model for predicting differences in risk for BRCA1 allele effect and location. Although the CIs around the OR estimates for this baseline model indicated a distinctly lower risk associated with BRCA1 alleles of severity score 2 compared to those with a score of 1 or 0 (both ORs = 11), the risk estimate itself falls into the range (OR > 4.0) considered high risk for breast cancer³⁹ and is thus unlikely to inform changes in clinical management for BRCA1 pathogenic variant heterozygotes in relation to breast cancer risk.

Table 4.

Risk of breast cancer according to allele severity score: Summary results from meta-analysis of three datasets

Gene	Allele severity option	Allele severity scoring description	Allele severity score	Total distinct variants (n)	Case Het (n)	Control Het (n)	OR (95% CI)	p
BRCA1	baseline	PTC (noNMD) score 2; PTC exon 10 score 1 (in-frame exon rescue); other PTC score 0; impacts start site score 0; splicing complete impact score 0 and partial impact score 1; missense in domain complete functional impact score 1 and partial or conflicting impact score 2	0	114	362	64	11.36 (8.51–15.17)	5.92 × 10−61
			1	205	531	124	11.06 (8.86–13.80)	1.44 × 10−100
			2	13	48	26	4.88 (2.89–8.26)	3.34 × 10−9
	A	as for baseline but (1) replace missense functional impact score with missense stability score (unstable score 1 and stable score 2)	0	114	362	64	11.36 (8.51–15.17)	5.92 × 10−61
			1	205	528	123	10.94 (8.75–13.67)	2.20 × 10−98
			2	13	51	27	5.46 (3.30–9.02)	3.87 × 10−11
	B	as for baseline but (1) replace missense functional impact score with missense stability score (unstable score 1 and stable score 2) and (2) apply score 1 for subset of splicing variants that lead to transcripts encoding a larger in-frame deletion expected to escape NMD	0	108	351	61	11.34 (8.44–15.23)	1.75 × 10−58
			1	209	527	125	10.87 (8.71–13.56)	5.26 × 10−99
			2	15	63	28	6.15 (3.78–9.99)	2.338 × 10−13
	C	as for baseline but (1) replace missense functional impact score with missense stability score (unstable score 1 and stable score 2); (2) upweight score for subset of splicing variants that lead to transcripts encoding a larger in-frame deletion expected to escape NMD; and (3) apply score 1 for PTC variants within BRCA1 exons 8 and 9 due to potential multi-exon in-frame rescue	0	106	350	60	11.47 (8.53–15.42)	1.23 × 10−58
			1	211	528	126	10.80 (8.66–13.47)	9.10 × 10−99
			2	15	63	28	6.15 (3.78–9.99)	2.33 × 10−13
	D	as for baseline but (1) replace missense functional impact score with missense stability score (unstable score 1 and stable score 2); (2) upweight score for subset of splicing variants that lead to transcripts encoding a larger in-frame deletion expected to escape NMD; (3) apply score 1 for PTC variants within BRCA1 exons 8 and 9 due to potential multi-exon in-frame rescue; and (4) upweight PTC variants in BRCA1 exon 10 due to additional molecular evidence for in-frame rescue	0	106	350	60	11.47 (8.53–15.42)	1.24 × 10−58
			1	34	106	17	13.98 (7.39–26.46)	5.23 × 10−16
			2	192	485	137	9.40 (7.60–11.61)	2.45 × 10−95
BRCA2	baseline	PTC (noNMD) score 2; PTC exon 10, 11, and 12 score 1 (exon rescue); other PTC score 0; splice site with regulatory impact score 1; splicing complete impact score 0 and partial impact score 1; missense outside domain (all with splice impact) score 0; missense in domain complete functional impact score 1 and partial or conflicting impact score 2	0	235	452	158	8.45 (6.85–10.43)	8.96 × 10−88
			1	391	1,014	585	5.33 (4.72–6.01)	2.07 × 10−163
			2	7	16	46	1.26 (0.58–2.72)	5.56 × 10−1
	A	as for baseline but (1) replace missense functional impact score with missense stability score (unstable score 1 and stable score 2)	0	235	452	158	8.45 (6.85–10.43)	8.96 × 10−88
			1	382	970	561	5.35 (4.73–6.05)	2.64 × 10−156
			2	16	60	70	2.64 (1.73–4.04)	6.97 × 10−6
	B	as for baseline but (1) replace missense functional impact score with missense stability score (unstable score 1 and stable score 2) and (2) apply score 1 for subset of splicing variants that lead to transcripts encoding a larger in-frame deletion expected to escape NMD	0	226	444	152	8.78 (7.08–10.88)	9.68 × 10−88
			1	390	978	567	5.30 (4.69–5.99)	3.75 × 10−156
			2	17	61	70	2.66 (1.74–4.07)	5.71 × 10−6
	C	as for baseline but (1) replace missense functional impact score with missense stability score (unstable score 1 and stable score 2); (2) apply score 1 for subset of splicing variants that lead to transcripts encoding a larger in-frame deletion expected to escape NMD; and (3) apply score 1 for PTC and missense variants within BRCA2 exons 4–7 due to potential multi-exon in-frame rescue	0	203	404	127	9.59 (7.64–12.05)	5.43 × 10−84
			1	413	1,018	592	5.27 (4.67–5.94)	2.97 × 10−160
			2	17	61	70	2.66 (1.74–4.07)	5.69 × 10−6
	D	as for baseline but (1) replace missense functional impact score with missense stability score (unstable score 1 and stable score 2); (2) apply score 1 for subset of splicing variants that lead to transcripts encoding a larger in-frame deletion expected to escape NMD; (3) apply score 1 for PTC and missense variants within BRCA2 exons 4–7 due to potential multi-exon rescue; and (4) apply score 1 for PTC in exon 14 due to exon size and potential to escape NMD; allele scoring method as for best-fitting BRCA2 FA genotype severity score model	0	186	384	121	9.37 (7.41–11.85)	6.65 × 10−78
			1	430	1,037	599	5.34 (4.74–6.02)	6.72 × 10−166
			2	17	61	70	2.66 (1.74–4.07)	5.69 × 10−6

Results for individual datasets BRIDGES, CARRIERS, and UK Biobank are provided in Tables S15 (BRCA1) and S16 (BRCA2). Total n: affected individuals = 96,691 and control subjects = 302,116. PTC, protein termination codon; NMD, nonsense-mediated decay; Het, heterozygote; OR, odds ratio; CI, confidence interval.

For BRCA2, all allele severity score options revealed differences in the magnitude of risk for alleles with a severity score of 0 (alleles annotated as having the most severe impact, OR range: 8.45–9.57) compared to those with a severity score of 1 (OR range: 5.30–5.34), with non-overlapping CIs for within-option comparisons. For alleles with a severity score of 2 (least severe impact), the risk estimate was below 3, ranging from a non-significant 1.3-fold risk (0.58–2.71) for baseline annotation to an OR of 2.64 (1.72–4.03) for option A or an OR of 2.66 (1.74–4.07) for all the remaining score options. Based on the continuous test (Table S16), there was no obviously better separation of effect for one scoring option method over another. For allele severity scoring option D, equivalent to the severity score model with the best fit for association with age at cancer onset in BRCA2 individuals with FA, the magnitude of breast cancer risk differed according to allele score, as follows: score of 0, OR of 9.37 (95% CI: 7.41–11.85); score of 1, OR of 5.34 (95% CI: 4.74–6.02); and score of 2, OR of 2.66 (95% CI: 1.74–4.07). Overall, these findings indicate that this allele severity scoring system, or perhaps an improved derivation thereof, could potentially be used as an indicator to highlight individual variants as suspected reduced-penetrance BRCA2 alleles, with implications for breast cancer risk management stratification of individuals heterozygous for a BRCA2 pathogenic variant.

Conclusions

Using a comprehensive data collection instrument to record features observed in individuals with FA due to BRCA1, BRCA2, or PALB2 pathogenic variants, our study results support the need to expand the list of FA-related features to include additional phenotypes. This includes the need for flexibility to incorporate additional HPO terms that span the grandparent-parent-child-grandchild hierarchy of existing FA-related terms (ORPHA:84) and a possible expansion to record features captured by multiple HPO terms unique to VACTERL-H (ORPHA:3412), or drawn from the literature, at a frequency ranging from 5% to 24%. Further, relationship mapping of an additional 178 phenotypic characteristics identified in individuals with FA suggested that approximately 50% of these features could potentially represent additional terms relevant for phenotypic diagnosis of FA in individuals with BRCA1, BRCA2, or PALB2 pathogenic variants. Additional study will be required to provide statistical evidence for this and also to assess if these additional terms might be relevant for phenotypic diagnosis of FA due to pathogenic variants in other genes. The hierarchical relationship of many of the characteristics observed in individuals with FA is unsurprising and partly reflects the use of multiple HPO terms that provide different levels of detail to describe a “single” given characteristic. It is important to note that ongoing updates to the HPO database can also impact relationship mapping over time. Our findings indicate that features related by close hierarchy to existing FA HPO terms should be permissible for assigning the PM3 code in variant interpretation and that “fuzzy matching”⁴⁰ should be considered to improve genetic diagnosis of FA within databases. It will also be beneficial to include a collection of data for the additional potentially FA-related terms in future research and clinical efforts, to enable independent validation as required to formally justify expansion of the ORPHA:84 HPO term list.

Although most individuals in our highly selected cohort presented with one or more recognized FA-related phenotypic features, our findings raise the importance of phenotyping for subtle features to better inform the use of recessive disease presentation (or lack thereof) for the classification of variants in BRCA1, BRCA2, or PALB2. Further, for individuals whose clinical features were documented through both previously published reports and prospective data collection, there was a significant difference in the number and characteristics of features reported, most notably for abnormalities of blood and blood-forming tissues. It is thus necessary to acknowledge that under-reporting of FA-related features may impact the assignment of the recessive presentation PM3 code for some variants, which is currently weighted based on a combination of chromosome breakage test results, FA physical features, and hematological and toxicity findings. We suggest that the use of a standardized data collection instrument, modeled on the one used here for our prospective data collection, would improve the reporting of FA-related features in both general clinical practice and publications.

Another important consideration for the PM3 code assignment following the current ClinGen ENIGMA BRCA1 and BRCA2 VCEP specifications is prenatal presentation. Five of the nine individuals with FA diagnosed prenatally were captured by the prospective data collection, perhaps reflecting increasing genetic analysis of suspected abnormalities detected during routine prenatal medical examination. Our data indicate that individuals with FA identified prenatally early in gestation are less likely to develop cancer and, from a practical perspective, are less likely to be examined for hematologic or chromosomal abnormalities. This raises the question of whether, for the purpose of variant interpretation, prenatal presentation could be considered as a severe clinical entity and, together with other known FA-related physical features, could perhaps be considered as sufficient to assign “recessive clinical features met” for individuals with bi-allelic BRCA1, BRCA2, or PALB2 variants.

Another component of our study was to assess the possibility that FA-related variants are more likely to be hypomorphic, as detected by experimental assays, and/or associated with reduced penetrance in heterozygote individuals. This is an important consideration since the ACMG/AMP classification system overall assumes a Mendelian high-risk disease model, and by inference, the PM3 code weight assumes that both variants observed in recessive presentation are associated with the dominant disease risk expected for the average pathogenic variant in the relevant gene. Of note, a recent publication reporting a framework to standardize interpretation and reporting of BRCA1 and BRCA2 reduced-penetrance pathogenic variants⁴¹ specifically denoted the identification of bi-allelic FA-affected individuals as evidence of a variant associated with reduced penetrance, together with repeated allele observations in compound heterozygote status in individuals with FA, and the potential for the allele to partially retain protein function. Our basic categorization of predicted and observed variant types and impact for all FA-related alleles showed that more than 60% of the distinct FA-related alleles (and more than 67% of allele observations across families) comprised traditional loss-of-function variant types. Consideration of currently collated functional assay results did not suggest an obvious enrichment of hypomorphic missense alleles in individuals with FA. However, genotype-phenotype correlation analysis, building on a previous report demonstrating an association of the BRCA2 allele severity score with age at cancer diagnosis in individuals with FA,²⁶ provided additional insight into allele features that may be used to predict cancer risk. The best-fitting model for predicting age of cancer diagnosis in BRCA2 individuals with FA was achieved by considering potential in-frame mRNA isoform rescue of PTC alleles, replacing “protein” functional results with bioinformatic prediction of protein stability, and annotating both the level and predicted/observed in-frame impact of alleles leading to splicing aberrations. Despite the highly significant association achieved by this model, noted examples of between-sibling differences in cancer diagnosis raise questions about the clinical utility of the scoring system to predict the risk of malignancy in BRCA2 individuals with FA at first diagnosis and invite further study to investigate possible explanations for between-sibling differences in presentation, e.g., genetic or environmental modifiers. It is interesting to note that within-family variability was only observed for families with a genotype severity score of at least 1, consistent with a testable hypothesis that presentation for “less severe” alleles would be more likely to be modified by other factors. Likewise, further study and much larger cohort sizes will be required to validate these findings and assess their relevance to predict cancer presentation in individuals with FA due to BRCA1 or PALB2 pathogenic variants.

Acknowledging such caveats, it is notable that the application of an allele severity scoring system to females heterozygous for BRCA1 or BRCA2 pathogenic variants revealed evidence for alignment of the allele severity score with the magnitude of breast cancer risk in the general population. While validation of these findings will be essential, the noted differences in breast cancer risk estimated by the BRCA2 allele severity score could potentially be used to inform differences in risk management for breast cancer. Further, consideration of the association of the allele severity score with the magnitude of risk for other cancer types, such as ovarian, prostate, and pancreatic cancer, will be critical for informing a comprehensive management strategy for individuals with cancer. It is relevant to note that previous segregation analyses comparing breast and ovarian cancer risk associated with BRCA1 or BRCA2 pathogenic PTC variants vs. missense variants reported evidence that BRCA1 and BRCA2 missense alleles overall confer a different breast cancer age-related risk profile, but no obvious differences in ovarian cancer risk profile, compared to BRCA1 or BRCA2 PTC variants.⁴² However, in the same study, penetrance analysis limited to 34 families with the BRCA2 c.7878G>C (p.Trp2626Cys) variant indicated that this FA-related allele (assigned a severity score of 2 in the final model) was associated with a reduced risk of both breast and ovarian cancer compared to the average PTC variant.⁴² Similarly, penetrance analysis focused on the BRCA1 c.5096G>A (p.Arg1699Gln) variant showed it to be associated with lower breast and ovarian cancer risk compared to the average high-risk pathogenic PTC variant.^35,36 Further, large-scale detailed penetrance analysis has revealed an atypical cancer risk profile for the FA-related allele BRCA1 c.5017_5019del (p.His1673del) (allele severity score of 2), with heterozygotes demonstrating increased risk of ovarian and uterine cancer, but not breast cancer, in families of Italian descent.⁴³ Nevertheless, these findings provide a starting point to explore the concept of the allele severity score as a predictor of the magnitude of cancer risk associated with BRCA1 or BRCA2 alleles considered pathogenic in the recessive vs. the dominant state. It will be important to consider alternative or additional experimental data or bioinformatic predictions for allele severity scoring permutations, including exploration of specific functional or mRNA assay methods to distinguish a pattern or range of impact enriched in variants with genetic evidence for reduced (or atypical) cancer risk compared to the average cancer risks associated with BRCA1 or BRCA2 pathogenic variants.

Importantly, the current breast cancer risk association findings presented here can be used to at least partly address the question of whether an FA clinical diagnosis in individuals can provide appropriately weighted evidence for the classification of BRCA1 or BRCA2 variants under the assumption of a high-risk model. BRCA1 individuals with FA were over-represented for alleles with an allele severity score of 2, but the magnitude of breast cancer risk for alleles in this category is sufficient to place them in the high-risk category. For BRCA2, 24% of all FA-related allele combinations included at least one allele with an impact severity score of 2, a category associated with a 2.6-fold moderate level of risk of breast cancer, for which clinical management recommendations are altered compared to classical high-risk BRCA2 variants.³⁹ Overall, this would suggest that an FA diagnosis does provide appropriate evidence toward pathogenicity for the majority of BRCA2 alleles detected in such individuals with FA. However, we suggest that that our findings may provide the impetus to introduce the allele severity score as an independent factor in future extended variant classification protocols, shifting the current binary pathogenic-benign classification paradigm to better discriminate between clinically actionable high-risk pathogenic variants, “reduced/low” penetrance pathogenic alleles, and variants for which clinical action is not justified based on that genetic information alone (including benign variants as well as “risk alleles” falling below clinically actionable thresholds).

Our study was not designed to address whether “lack of FA phenotype” is a reliable negative predictor of BRCA1, BRCA2, or PALB2 variant pathogenicity, specifically in relation to alleles with reduced (or atypical) penetrance compared to the average pathogenic variant in that gene. However, we note that there is emerging evidence that FA phenotype, assessed against any of the three broad measures currently used in VCEP recommendations, can be absent from individuals who are compound heterozygous or bi-allelic for BRCA1 variants with measurable deleterious impact on function that could confer a clinically significant magnitude of cancer risk. Lack of FA phenotype has been reported for individuals bi-allelic for the BRCA1 c.4096+3A>G allele,^44,45 a variant that upregulates expression of the BRCA1 Δ10q in-frame deletion isoform (termed Δ11q using BRCA1 legacy exon numbering) and has been shown from case-control analysis to be associated with a 3-fold risk of breast cancer and an 8-fold risk of ovarian cancer.⁴⁶ As another example, a male without an FA phenotype or BRCA1-related cancer was identified to be compound heterozygous for a BRCA1 frameshift variant and an in-frame duplication with an impact of transcription activation equivalent to that reported for the known reduced-penetrance variant BRCA1 c.5096G>A (p.Arg1699Gln).³⁸

Together, the results arising from this study have implications for the application of BRCA1, BRCA2, and PALB2 genetic variant information in clinical practice. Overall, our findings have expanded the list of HPO terms to be considered in the clinical diagnosis of BRCA1-, BRCA2-, or PALB2-related FA and provided interesting insights into the future BRCA1 and BRCA2 genotype-phenotype correlation studies that consider variant type, predicted/observed molecular impact, location of variant, and also possible rescue due to in-frame part, single, or multi-exon skipping. It will be of interest to conduct large-scale, well-powered case-control studies to assess if a similar allele severity scoring approach may provide insight into differences in risk of other (non-breast) cancer types known to be associated with BRCA1 or BRCA2 pathogenic variants and potential differences in cancer risk associated with pathogenic variants in PALB2 and other cancer predisposition genes.

Data and code availability

There are restrictions on the availability of individual-level HPO terms for individuals with FA due to privacy reasons. Such information can be made available from the corresponding author upon request and subject to relevant ethical approvals.

Source data for case-control analysis may be accessed as follows: BRIDGES data, upon request through the BCAC Data Access Coordinating Committee (BCAC@medschl.cam.ac.uk); UKB data, via the UKB Access Management Team (access@ukbiobank.ac.uk); and CARRIERS, genotype data through dbGAP: phs002820.v1 and phenotype data upon request through the CARRIERS Coordinating Committee (couch.fergus@mayo.edu).

Acknowledgments

We are indebted to the generosity of the individuals with FA and their families, as well as their treating physicians, for participation in research supporting this study. We thank all the individuals who took part in or enabled the cohort studies informing the breast cancer case-control genetic studies. Additional information relating to acknowledgments and funding sources is provided in the supplemental information.

Author contributions

Conceptualization, S.E.J., K.T., J.N., M.T., A.N., E.G.-G., and A.B.S.; methodology, S.E.J., M.T.P., K.M., M.Z., J.N., M.T., A.N., E.G.-G., and A.B.S.; validation, A.B.S.; formal analysis, S.E.J., E.T., K.M., M.Z., and A.B.S.; investigation, S.E.J., E.T., M.T.P., D.M.C., A.L.D., T.B., R.O.R., C.P.K., R.K., L.J.M., N.G., M.E.R., T.P., J.S., R.P., B.R.V., M.G., K.N.M., K.N., S.D., M.O.F.-R., S.F., B.G.-d.-T., M.J., S.L., M.M., K.P., R.R., S.R., L.S.B., S.E.L.T., K.T., A.D.A., M.I.C., J.A.K., M.L.M., R.T., J.E.W., M.W., A.J.M.D.., P.J., T.V.O.H., D.F.E., A.S., J.N., M.T., A.N., E.G.-G., and A.B.S.; resources, K.M., M.Z., T.B., R.O.R., C.P.K., R.K., L.J.M., N.G., M.E.R., T.P., J.S., R.P., B.R.V., M.G., K.N.M., K.N., S.D., M.O.F.-R., S.F., B.G.-d.-T., M.J., S.L., M.M., K.P., R.R., S.R., L.S.B., S.E.L.T., A.D.A., M.I.C., J.A.K., M.L.M., R.T., J.E.W., M.W., N.J.B., C.H., J.N.W., J.H., N.R., L.R.-M., M.d.l.H., L.C.W., D.F.E., S.K.S., F.C., A.S., A.N., E.G.-G., and A.B.S.; data curation, S.E.J., E.T., M.T.P., D.M.C., A.L.D., J.H., N.R., L.R.-M., M.d.l.H., M.P.G.V., L.C.W., S.K.S., F.C., and A.B.S.; visualization, S.E.J., K.M., M.Z., D.M.C., A.L.D., A.J.M.D., J.N., M.T., E.G.-G., and A.B.S.; supervision, K.M., A.S., J.N., M.T., E.G.-G., and A.B.S.; project administration, S.E.J. and A.B.S.; funding acquisition, A.B.S.; writing – original draft, S.E.J. and A.B.S.; writing – review & editing, all authors.

Declaration of interests

J.S. and J.N. have received research funding from pharma and biotech companies unrelated to this research. T.V.O.H. has received lecture honoraria from AstraZeneca. K.T. has received reimbursement from Merck Sharp and Dolme for work related to von Hippel-Lindau disease.

Published: October 30, 2025

Web resources

• ClinGen Variant Curation Expert Panel (VCEP), https://cspec.genome.network/cspec/ui/svi/

• ClinicalTrials.gov, https://clinicaltrials.gov

• dbGAP, https://dbgap.ncbi.nlm.nih.gov/beta/home

• GenBank, https://www.ncbi.nlm.nih.gov/genbank/

• HPO project, https://hpo.jax.org/

• OMIM, https://www.omim.org

• Orphanet Rare Disease Database, https://www.orpha.net/en/disease.